“If you want to win a race condition, you best cheat”.
The following article demonstrates a fascinating method of exploiting Local File Inclusion vulnerabilities first thought up of by the guy who runs Gynvael.Coldwind.pl . It is further documented at Insomnia Security
To start, when you are exploiting a Local File Inclusion, in order to gain remote code execution you must “write” some PHP code somewhere on disc that you can “include”. Normally this means injecting into logfiles, or the /proc/self/environ interface.
In the event you cannot locate the logfiles, or access them, most people simply give up and claim “Yep, it is unexploitable”.
This is not true.
When you upload files via HTTP POST to a webserver running PHP, during the upload it creates a tempfile for the files data, normally in /tmp. This file is “randomly named” (I do not trust the PRNG PHP uses), and is stored for the duration of the upload.
Now, it just so happens that it will create this file irrespective of whether or not you are legitimately uploading (i.e. the webapp is accepting an upload) or not, and the file is deleted when the “upload” finishes.
Because these files are randomly named, and there are no wildcards that we can use on Linux (more on this later!), you might think “So, how does this help me at all?”.
Normally, you would be right. Unless you find a flaw in the PRNG that allows you to predict the “dummy name” of the file, this is not so useful to you. What we need is an information disclosure bug that displays the PHP variables at that particular time…
… Enter PHPinfo(). The phpinfo() call, often found in /phpinfo.php or other such files (often forgotten and not removed on webservers), displays the PHP variables as they are at that point in time… Including the “Dummy Filename” that is currently in use.
So, if we are to send a file upload, containing our evil PHP code, and then include() it (via the LFI), using the path given to us by the PHPinfo page, we gain code execution on the remote server.
Yes, but not quite.
The file exists there for mere fractions of a second, so we lengthen the duration of the upload (we upload to the phpinfo file) by padding it with trash data. The insomniasec paper (linked at bottom) explains this better than I will, so I advise reading it! This is a race condition, so to further enhance our chances, we use multiple threads and multiple attempts. One of them will succeed, normally within short order (averages at 30 seconds, see the demo video!). If you have LFI and a phpinfo() page, you WILL gain code execution on the server.
The PHP code you “upload” and execute via include() effectively acts as a “Dropper”, dropping some malicious code that is stored in /tmp with a filename you have set. You then simply include() this dropped file via the LFI bug to execute code, as per normal LFI.
The InsomniaSec guys released, along with their paper, a demo script to show how it works. imax, the mantainer of Fimap (which I also develop) released a Fimap plugin that exploits this vulnerability, and that is what I am about to demonstrate. imax’s code drops a file named “eggshell” which is essentially a stripped down version of the Fimap payload.
The Fimap module is extremely reliable, giving me shells in short order. I have never had an issue with it to date, and normally on remote boxes it takes a minute or so to pop a shell.
Next week I will be releasing a demo of doing this against Windows targets without the lovely PHPinfo bug. A much “truer” race condition